For the past million years, Earth's climate has repeatedly fallen into ice ages with a periodicity of ∼100 000 years . How glacial inceptions occur is still a largely unresolved problem in climate science . The leading hypothesis for the cause of glacial inception is that orbital variations cause cooler summers leading to less melting, although previous studies found that additional amplifying feedbacks beyond the ice–albedo feedback are necessary to explain inception. Possible amplifiers of orbital forcing may be divided into large-scale non-local feedbacks, including ocean temperature and planetary-scale circulation changes, and local feedbacks associated with greater height and extent of mountain glaciers , regional circulation changes, clouds, and more. In this study, we seek to understand specifically the role of local ice–climate feedbacks in glacial inception using a regional climate model asynchronously coupled to a simple ice sheet model. Our regional model is centered over Baffin Island – the likely location of the last glacial inception .

Glacial inception of the Laurentide ice sheet is thought to have progressed fairly rapidly from Baffin Island in eastern Canada , with the associated drop in sea level estimated to be as large as 50 m in 10 kyrs . The Laurentide ice sheet has been identified as the primary initial cause of the drop in sea level , and some ice modeling studies have shown rapid inception within a few thousand years . However, this rapid growth required extreme precipitation increases or excessive climate cooling every thousand years, without any explicit explanation. A more modest sea level drop of 10–20 m in 10 kyrs has also been posited , which indicates a slower ice sheet growth rate and possibly ice growth relegated to Arctic Canadian islands .

Global climate models (GCMs) have been used to simulate glacial inception, focusing on large-scale feedbacks. They often find that snow does not accumulate in unglaciated areas or that glaciers do not spread into unglaciated areas. One reason GCMs have difficulty simulating the glacial inception is their coarse horizontal resolution, which lowers the topographic features dramatically . Snow then does not accumulate because the region is too warm due to artificially low elevations . Though topography is important , other aspects of the climate system may also be misrepresented in GCMs, like ocean feedbacks , dust , sea ice , vegetation , and meteorology . Increasing the model resolution has led to moderate improvements in the simulation of glacial inception , through the occurrence of areas of net snow accumulation, which can progressively expand as the climate cools. Given that GCMs had such difficulties in simulating glacial inception, our study focuses on the importance of regional feedbacks over Baffin Island.

While Milankovitch forcing explains some summer cooling, the large-scale circulation is an important driver of heat and moisture fluxes to inception areas. In the current climatology, large-scale cyclonic circulation over Baffin Bay is associated with cool summers , which may be all that is necessary for inception. However, and also noted increased storms and poleward moisture fluxes during inception. showed that, in addition to favorable circulation and temperature, snow could accumulate in new areas only after artificial large-scale cooling was prescribed. The following question remains: how does this cooling occur? One possible answer is the interaction of ice sheets and the atmosphere.

As the ice sheet grows, it influences the atmosphere due to albedo and surface elevation changes. The regional circulation can also be diverted by the changing topography. Anticyclones have been noted to occur over mountains , but ultimately the effects of mountains on the regional climate and large-scale circulation depend on the mean overlying flow and mountain size . Looking at large-scale ice growth in the context of glacial inception, argued using a simple model that slight changes in topography can influence the stationary waves, significantly changing the circulation and leading to colder temperatures. The topography and geographic location may be critical to the glacial inception process , but previous studies did not consider the effect of ice-growth-induced regional topography changes on Baffin Island.

To capture the interaction of ice and the atmosphere, climate models of intermediate complexity have been combined with ice sheet models. This type of coupled ice–climate model allows thousands of years of ice growth due to changes in insolation and CO2 to be simulated. Often, however, these models glaciate Alaska and western Canada very rapidly , which is not consistent with the geological record showing inception occurred first over Baffin Island in eastern Canada . Ice sheet models and GCM studies have a similar problem of not agreeing with the geological record. Combining GCMs and ice sheet modeling may allow for progress in understanding ice–atmosphere interactions , but the smoothed topography and other biases in these coarse-resolution climate models often need to be artificially corrected to obtain an accurate regional climate. Furthermore, mountain ice caps of the Canadian Arctic archipelago in the present climate are too small to be modeled by large-scale ice models – hence our use of a higher-resolution regional ice model.

The interaction of ice and atmosphere is a critical component of glacial inception, often left out of climate models. For instance, focused on a single ice cap on Baffin Island using a high-resolution cloud-permitting regional climate model without an ice flow model. That study found that ice would accumulate on the mountain glaciers of Baffin Island. The 4 km resolution of that study captured both the ablation and accumulation zones of mountain glaciers. By integrating the mass balance from the top of the ice cap downslope, implicitly represented the effects of ice flow. As a result, the Penny Ice Cap was deduced to grow outside of its present-day bounds due to decreased insolation and a sequence of cold summers and wet winters. Consistent with previous GCM studies, no snow accumulated in unglaciated areas.

Thus, the purpose of the current paper is to study the role of regional feedbacks in the inception over the major ice caps of Baffin Island, including the role of ice flow. By prescribing boundary conditions from the modern climate, we do not consider any role of non-local ice sheet growth in modifying the planetary wave pattern, which could aid or hinder inception on Baffin Island. This is consistent both with other studies and with our intention to study the effect of regional feedbacks. Similarly, we neglect the role of sea surface temperature (SST) changes, as there is debate over warm or cold oceans during inception . We choose to keep modern SSTs and identify whether inception may be described as a local process over Baffin Island, driven by Milankovitch forcing. Regional ice–albedo and height–mass balance feedbacks are expected to dominate at least some of the planetary-scale feedbacks at the very beginning of the inception process. We drive the model with boundary conditions corresponding to wet and cool summers from reanalysis, ideal for inception but not extremely uncommon during present-day climate. Such conditions may be the result of SST forcing that is not explicitly simulated, so in that sense we do include some SST effects. Our regional model configuration could also be driven by the boundary conditions from a GCM simulation of glacial inception, which would incorporate some planetary-scale feedbacks. However, given the inability of such models to simulate the inception and often the lack of substantial circulation changes over Baffin Island , we choose not to take that route and focus solely on regional feedbacks.

We investigate the feedbacks between a growing ice sheet and the atmosphere at higher spatial resolution than can be simulated with GCMs, using an shallow ice approximation ice flow model asynchronously coupled with the Weather Research and Forecasting model WRF;. We find that the change to ice sheet topography induces an anticyclonic atmospheric response over Baffin Island. This anticyclone causes a flow of warm air over the Baffin Island and Hudson Bay area, increasing ablation, stagnating further ice sheet growth, and thus indicating a negative feedback on inception at the regional scale.

The atmospheric and ice models used for the asynchronous coupling experiment are presented in Sect. . We present the results in Sect. , beginning with the mass balances and resulting ice thickness of our atmosphere–ice iterations. We then investigate the regional climate and local circulation impacts of expanding ice cover and increasing ice elevation. Sensitivity simulations exploring the importance of topography and insolation are also presented. Finally, the sensitivity of ice flow to ice viscosity and mass balance are examined. We conclude with a discussion in Sect. , including the limitations of this study and its implications.

Our goal in this study is to better understand the interaction between changing ice cover and the regional climate of Baffin Island and the feedback on the regional circulation over Canada. We therefore use a nested configuration of the WRF model with the outer domain covering all of North America with a 100 km resolution and the inner domain covering Baffin Island with a 20 km resolution, encompassing both the Penny and Barnes ice caps. To better understand the growth of these mountain glaciers during inception, we must simulate ice flow in the inner 20 km resolution domain, but WRF does not contain an ice flow model, which means the ice expansion must be calculated offline. Based off of , we developed a simple ice flow model that reduces the 3-D Stokes flow to two dimensions using the shallow ice approximation. The basal sliding velocity is assumed to be 0. This approximation results in a non-linear diffusion equation that represents ice flow by evolving ice thickness H in time based on horizontal gradients of surface elevation H*, along with surface mass balance G: ∂H∂t=∇⋅D∇H*+G.

In Eq. (), the yearly surface mass balance G is calculated from WRF simulations and assumes that the density of ice is 900 kg m-3. The surface elevation H* and ice thickness H are linked by H*=B+H, where B is the basal topography. The non-linear shear thinning of ice flow is represented by a diffusivity, D, which depends on both the ice thickness and surface elevation gradients: D=AHm+2∂H*∂x2+∂H*∂y2(m-1)/2.

The parameter m for all model simulations is set to 3, which is a typical value for realistic ice simulations . The coefficient A in Eq. () is the creep parameter that influences the viscosity of the ice based on the temperature of the ice; i.e., the colder ice flows more slowly, while warmer ice flows more quickly. We use a creep parameter A=3.5×10-25 Pa-3 s-1 from . More details on this choice of creep parameter can be found in Appendix A, which also contains the details of the numerical scheme used to integrate this ice model in time.

The ice model requires the specification of initial surface elevation (H*), initial ice thickness (H), basal topography (B) that is time-independent, and mass balance (G) as an input for each time step. The ice thickness (H) is adapted from the present-day ice cover on Baffin Island from the NASA Operation IceBridge mission , while H* and G are taken from our WRF simulations. The grid size and time step are also user specified and chosen to be 20 km (same as WRF inner domain) and 1 year, respectively. With these atmospheric and ice models, we perform iterations, running WRF, using the mass balance calculated by WRF to force the ice model, updating the ice thickness and extent calculated by the ice model, and then running WRF again with the newly calculated ice cover as an input. This amounts to an asynchronous coupling of the two models, where a WRF run followed by the ice model is considered an iteration.

WRF , version 3.7, is a compressible, non-hydrostatic model, which has been shown to provide a reasonable simulation of mountainous and Arctic regions . Reanalysis data from the European Centre for Medium-Range Weather Forecasts ECMWF; are used for setting the meteorological boundary and initial conditions, including sea surface temperatures and sea ice extent. also noted that the interim reanalysis (ERA-Interim) is a suitable choice of boundary conditions for WRF in the Arctic.

We use the WRF single-moment six-class microphysics scheme, which allows for ice, snow, and graupel processes . The community Noah land surface model with multiparameterization options (Noah-MP) is used for the land surface model. It includes four layers of snow and allows for refreezing, which is an important process on the Penny Ice Cap according to historical trends . Noah-MP caps the snow water content at 2000 kg m-2 by default, but we remove this limit. The radiation scheme is the Rapid Radiative Transfer Model for GCM applications RRTMG;, which has been found to have minimal bias in WRF . CO2 is set to 290 ppm for a glacial inception scenario . The boundary layer is simulated with the Mellor–Yamada–Janjic scheme MYJ;. Lateral boundary conditions are prescribed at the edges of the outer domain, yet no nudging to observations is used in the domain interior, as we want to allow the atmosphere to respond to land surface changes. Such spectral nudging is often used in regional models to eliminate artificial circulation patterns that develop in spite of the prescribed boundary conditions . Comparing a present-day control run to ECMWF, we find (Fig. S1 in the Supplement) non-negligible deviations in geopotential height. Yet we show below that the anomalous circulation that develops in response to ice growth in our simulations is likely unrelated to these anomalies, as they develop gradually in response to the growing ice height. Note that these physics are the same as those used in our previous paper , except now we use the Grell convective scheme , which has been shown to give good results in the present day and the Last Glacial Maximum (LGM) Arctic climates.

We use realistic present-day land types from the WRF database as initial conditions for the ice extent on Baffin Island. Over these land grid points classified as ice, a snow water equivalent (SWE) of 5000 kg m-2 is prescribed – sufficient for maintaining snow cover for the duration of our runs and allowing us to capture a representative yearly mass balance for the chosen climatology. Greenland is initialized with present-day ice cover and elevation based on studies finding little influence of Greenland on the build up of the Laurentide ice sheet .

We first run a present-day meteorology simulation forced by ECMWF boundary conditions, representative of average and cold atmospheric conditions, also simulated by . With the above WRF configuration, we find a reasonable reproduction of the Arctic atmospheric state when we look at the full fields (Fig. S1). While there are deviations in geopotential and temperature between WRF and ECMWF, the differences in geopotential have been seen in other Arctic studies . The temperature bias occurs from differences in the snow cover between the models. For instance, ECMWF has been noted to have too large of a snow-covered area on coasts , promoting cooling on Baffin Island which is mostly coast at the reanalysis resolution. Also, June is the cause of the large temperature bias because ECMWF still has snow on the ground, associated with delayed snowmelt , while the snow in WRF melts earlier. July and August only differ in temperature by about 1 ∘C, when the snow cover in ECMWF and WRF are more similar. With the warm temperature bias, WRF will likely not cause an artificial inception, which making it more suitable than models with a cold bias. Furthermore, ECMWF is known to have a cold bias and a significant low geopotential height in the central Arctic of the same magnitude as the deviation we see in Fig. S1. Therefore, with WRF having the opposite biases, the Arctic climate simulated is not abnormal. Our WRF simulation matches observations of precipitation on the coast of Baffin Island, important for the surface mass balance, of about 300 mm per year . In any case, the warm bias in our WRF control run does not prevent the circulation from bringing relatively cold and wet conditions to Baffin Island . Previous works also proceeded to look at climate sensitivity by comparing model experiments in spite of biases in the control run , as we do here, and we therefore do not believe the biases would have a significant impact on the sensitivity results here.

For our main set of glacial inception simulations, the incoming solar radiation is set to values consistent with 115 kya; the process of modifying the insolation in WRF based on the orbital parameters is detailed in the Appendix of . The conditions most conducive to snow accumulation are from October 1985 to October 1986. These cold meteorology boundary conditions caused the present-day ice cover to expand when combined with 115 kya insolation in the study of . Figure S3 in the Supplement shows cumulative mass balance for present-day average conditions, present-day cold meteorology conditions, and cold meteorology plus 115 kya orbital forcing. The combination of moderately cool temperatures and reduced summer insolation leads to net snow accumulation, hinting that inception might be viable given sufficiently positive regional feedbacks. A longer averaging of the forcing, as obtained from a climate model, would have some advantages. However, given the persistent difficulties in simulating glacial inception using such GCMs, our goal here is to minimize the introduction of biases from global climate simulations. We therefore prefer to choose the necessary cold and wet meteorology from the observed modern climate (reanalysis). We cannot explain how these ideal circulation patterns occur during inception, but a future study with a GCM might carefully explore this. We do note that forcing with a single-year meteorology may bias the ice growth patterns if that year is not representative of longer-term cold conditions.

In the context of glacial inception, expanding ice cover interacts with the atmosphere, causing changes to both the regional climate and atmospheric circulation near Baffin Island. We first examine the progression of the WRF–ice model iterations, looking at the expansion of ice cover and the resultant changes in mass balance. Then, in Sect. , the changes in mass balance are explained by exploring changes in temperature, ablation, precipitation, and cloud cover. In Sect. , we take a closer look at the regional circulation. Additional sensitivity simulations with the ice sheet model further clarify climate factors that could lead to Baffin Island glaciation in Sect. .

The progression of WRF–ice model iterations (5 years of WRF and 500 years of the ice model) is laid out in Fig. . The ice thickness calculated by the ice model and then used in WRF is depicted in the left column, and the resulting annual mass balance from WRF in the right column. Also included in the right column is the spatial average of mass balance on the ice cap (defined as all grid cells with ice land cover type). Starting with the top left, we show the ice thickness resulting from running the ice model for 500 years using the Iteration 1 WRF mass balance (Fig. 1b, c). This new ice thickness, ice extent, and surface elevation computed by the ice model are used as inputs to the next WRF simulation. Then, WRF is run for 5 years resulting in the mass balance presented in the right column. After the second iteration of WRF, we bin the mass balance by elevation (as seen in Fig. 1c) and force the ice model with this new updated mass balance. A new ice cover is computed (Fig. 2a, second row) and used as an input in WRF. Subsequent iterations pictured follow the same procedure (going down Fig. 2, left to right).

The second iteration has the greatest ice extent, which is due to the large positive Iteration 1 mass balance. In the third iteration, the ice retreats from its margins, though there is still growth in the highest parts of the domain. Additional iterations show that the ice cap coalesces and increases in thickness to nearly 2000 m but expands slowly into new areas. After 5 kyrs, the ice accumulation on Baffin Island would only lower global sea level by 0.6 m. Although the average mass balance does generally increase with time, the rest of our analysis focuses on why the ice does not expand quickly.

The binned mass balance used in the ice model is plotted in Fig.  with annual mass balance on the y axis and surface elevation on the x axis. The Iteration 1 WRF run is the “+” blue line with the most positive mass balance, particularly at the lowest and highest elevations. In subsequent WRF runs, the mass balance up to 1000 m becomes more negative at low elevations, which we find surprising. One might expect the equilibrium line to shift downslope as a result of regional cooling from the initial expansion of the ice cover, leading to a more positive mass balance. However, plotting the binned mass balance for all iterations shows a general trend of increasingly more negative mass balances at low elevations, with the equilibrium line altitude shifting upwards by several hundred meters over the course of 10 iterations. As a result, the ice caps coalesce and thicken at high elevations but stagnate and fail to expand horizontally. The mass balance at highest elevations from all iterations also illustrates desertification or the mass balance becoming less positive over time as the ice cap increases in height and its top surface flattens.

Lateral expansion is slow because the lower terrain on Baffin Island is not strongly sloped. Thus, ice must build up to a substantial height before the spread of ice outweighs ablation at the ice edge. For example, with flat basal topography, calculations based on Eqs. () and () suggest that a 0.5 m yr1 of ice thickness increase due to ice flow in a grid cell with no ice requires an adjacent grid cell ice thickness around 500 m. This approximate ice thickness on the margins occurs in our simulations as depicted in Fig. .

In this paper, we have attempted to simulate the last glacial inception beginning at 115 kya with a regional high-resolution configuration of WRF, coupled asynchronously with a simple ice flow model. A net positive mass balance occurs in the first WRF iteration simulation driven by 115 kya insolation, cold present-day (1986) boundary conditions, and present-day ice cover. If the mass balance from this first iteration was maintained, an ice sheet could cover all of Baffin Island after 10 kyr. However, updating the ice cover and surface elevation in WRF creates a negative feedback from changes in the large-scale circulation: summer temperatures over Baffin Island increase due to a regional anticyclonic circulation. This causes the mass balance over Baffin Island to become more negative at low elevations, stagnating further ice expansion in the ice flow model. This is reminiscent of the findings of and , who noted the continued growth of the ice sheet is slowed by changes in large-scale circulation that arise from atmosphere–topography interactions. While the studies of and found the negative feedback slowed ice growth mostly when the ice sheet was fairly developed, of a continental scale, the negative feedback analyzed here occurs upon inception, as soon as the ice caps over Baffin Island begin to grow. Previous studies could not resolve this feedback with their coarse-resolution GCMs, while our coupled ice sheet – regional atmospheric model enables us to examine it.

The flow over Baffin Island driven by the 1986 boundary conditions is consistently from the northwest during the summer, with weak speeds of about 6 m s-1 at 500 mbar. The increase in the ice surface elevation after the first iterations of the coupled model causes an anticyclonic response of about 1 m s-1 to develop to the south of Baffin Island, while a cyclone develops in the north, diverting the winds and bringing in warmer air from the south. This leads to net ablation and arrests the glaciation process. This circulation response is likely dependent on the initial flow (direction and speed) and topographic height . Thus, a different meteorology choice might lead to sustained accumulation. identified exceptionally cold conditions on Baffin Island and their varying large-scale circulation patterns. Generally, the cold summer circulation flows from the northwest or north–northwest. We examine the role of meteorology directly using boundary conditions that differ from those used in our cold meteorology simulations. For this purpose, WRF is driven using 1979 boundary conditions, which is an average year in regards to temperature and precipitation, and the flow over Baffin Island generally comes from the west. WRF simulates a negative mass balance for this year, and the ice sheet retreats to an elevation above 1400 m when the ice model is forced with this mass balance for 500 years. In contrast, the ice cap from our average 1979 meteorology at a 4 km resolution was in approximate equilibrium, leaning towards slight expansion . Net ablation occurs in the 20 km resolution because of the lower topography. It is possible that during inception the circulation pattern was different and allowed for both the initial ice expansion and continued cold conditions. WRF did not simulate such a pattern in this study, nor did the few “cold” meteorological conditions in the ERA-Interim reanalysis.

We find significant warming on Baffin Island and the surrounding Hudson Bay region in response to the initial ice growth due to anticyclonic flow. The formation of anticyclones over expanding ice sheets has been noted to allow cold northerly winds to flow over the ice sheet, enhancing inception . In particular, small ice sheets cause a wave response that enhances accumulation, while stationary waves induce less cooling as ice sheets increase in size . In contrast, the anticyclone in our simulations forms south of the ice growing on Baffin Island, bringing significant warming even over these relatively small mountain glaciers. Similar to our study, found the Hudson Bay warmed in their coupled atmosphere–ice simulations due to anticyclonic flow. also found warming on the margins from anticyclones and reduced cloudiness. In both of these studies, though, the warming only slowed growth; the regional cooling due to the growing ice sheet kept the ice advancing. We find the heat fluxes into the region by the developing anticyclone circulation to completely overpower any regional cooling from the snow–albedo feedback, stagnating the ice sheet growth. We do observe the development of katabatic winds in response to ice sheet growth in our model, possibly contributing to the drying out of the region and decreasing cloud cover, but these winds do not cool the area outside of the glaciers sufficiently to allow for inception.

We do not find a positive feedback that would lead to a more positive mass balance due to the initial ice growth. However, some of our assumptions and modeling choices could be preventing such a positive feedback. For instance, our WRF simulations prescribe the SSTs and sea ice to those of 1986, but during glacial inception the SST could decrease and sea ice extent grow larger . Allowing the SST to cool has been shown to induce cooling over land, even if the atmosphere is relatively warm , while noted that storm tracks change with colder SSTs in a manner conducive for inception. In contrast, and argued that the oceans change slower than the atmosphere, so the oceans may remain warmer and ice free at the beginning of inception. argued that SSTs remained warm until 105 kya, when substantial ice sheets had formed. noted that the northern Atlantic remained close to modern-day temperatures through 115 kya until glacial stage 5d around 110 kya. argued that observations support a warm ocean, similar to an interglacial, during inception, though their focus is on the Southern Hemisphere. Some past GCM studies yielding some degree of glacial inception have also found Atlantic Ocean temperatures to remain similar to the present day .

Oceans may also affect moisture availability, not just heat fluxes. The amount of sea ice can impact accumulation over the ice sheet due to decreasing atmospheric moisture and evaporation from ocean, according to the temperature–precipitation feedback . Some studies argued that the cooling during inception did not cause sea ice to extend outside of Baffin Bay until 80 kya , meaning the Atlantic remained ice free, which was conducive to high precipitation. Both and found that including ocean feedbacks caused increased moisture fluxes conducive to inception, while and noted that storm frequency increased. Including ocean feedbacks could lead to increased moisture fluxes and thus precipitation, or it could lead to cooler temperatures decreasing melting, which could be more important for inception. With present-day fluxes prescribed at the boundaries, and the focus of this study being regional feedbacks, we keep SSTs at modern values for consistency, and effectively investigate the “warm” interglacial temperature ocean scenario, which does have support in observations .

We fix the CO2 concentration to 290 ppm, although CO2 decreased during inception. This decrease has been shown to influence the accumulation of Northern Hemisphere ice sheets , accounting for up to 50 % more ice . We also do not allow for dynamic vegetation, but on Baffin Island the tree canopy does not exist. Most of these studies dealing with dynamic vegetation refer to the treeline moving southward during glacial inception , enhancing ice growth, but on Baffin Island the vegetation is mostly tundra and small shrubs, which are not an impediment to snow accumulation as trees would be . However, the absence of dynamic vegetation means we cannot explore the cooling effects of albedo changes outside of Baffin Island due to vegetation changes. The southward treeline trend and the subsequent albedo impact, though, have been shown to occur later in the inception process , justifying our neglect of this feedback.

We choose to drive our model with meteorology conducive to a glaciation inception, and force the model with anomalously high precipitation and cool temperatures over Baffin Island in 1986. We are assuming that the planetary waves' response to the growing Laurentide ice sheet was not significant until the size of the Laurentide was substantial, as has been found in other modeling studies of glacial inception . The regional flow induced by these boundary conditions in conjunction with ice growth causes the negative feedback discussed above. The strength of this negative regional feedback that prevents inception from proceeding suggests that a reorganization of the planetary-scale Northern Hemisphere circulation could occur as part of the inception process and counter the regional negative feedback. We cannot capture such a change in the general circulation with our regional model and prescribed boundary conditions, indicating a study of non-local processes could complement this study of local Baffin Island feedbacks. For instance, the Laurentide ice sheet and Fennoscandian ice sheet in Europe are thought to have begun growing at the same time, though the Laurentide increased in volume more rapidly . The European ice sheet may have influenced the formation of the Laurentide , but such a feedback is missing in our regional model. noted that changes in the stationary wave pattern would affect the downstream climate.

We limit our study to Baffin Island based on geological records of it being the nucleation of the Laurentide, but regional modeling of other Canadian mountain ranges could help identify how easily other areas glaciated, contributing to sea level drop. Growth of ice at high elevations simultaneously throughout eastern Canada is not impossible, which could all coalesce into the Laurentide. For instance, various ice sheet models have shown ice beginning to accumulate all over Canada . While many of these ice modes do not match geological records because they simulate the growth of ice in the west, ice growth on many of the more eastern mountain ranges could perhaps impact the large-scale circulation in a manner conducive to inception.

Though we do find net accumulation in our Iteration 1 simulation using a 20 km resolution, there are inherent issues that come from this resolution. The topography of Baffin Island is very rough but is greatly smoothed by the 20 km resolution. More importantly, the highest elevation peaks with 20 km resolution are ∼400 m below the peaks found in realistic Baffin Island topography. Our simulation with average meteorology demonstrates that resolution increases can easily switch the mass balance from net accumulation to net ablation. For cold meteorology, the resulting mass balance is similar to what we found with a 4 km resolution .

To get an average mass balance to drive the ice sheet model, WRF is run for 5 years; we use the average mass balance from the last 4 years. We find that the land model required a 1-year spin-up time, although often a spin-up time of more than a year is recommended . We tested running the model for longer than 10 years, but the simulations did not deviate much from years 2–5 of our simulations. showed that multiple ensemble members of an atmospheric model simulation can give different results on a timescale up to 50 years, but this may not be an issue here because we are driving our regional model with ideal inception boundary conditions that constrain the synoptic and planetary-scale flows .

Using the shallow ice approximation neglects important ice dynamics; in particular, it neglects lateral stresses and does not allow for the development of ice streams, which cause rapid ice flow. It has also been suggested that a finer grid size in the ice model could make for more accurate flow, as coarser resolutions have been shown to halt ice flow . As noted above, lateral expansion only occurs when substantial ice thickness exists in an adjacent grid cell. However, we find that simulations with a 4 km resolution (Fig. S6 in the Supplement) yield similar ice flow patterns to the 20 km results we present here. We are neglecting basal sliding that could occur as the ice base melts due to geothermal heat flux , but the initiating ice sheet was likely frozen at the bed , which makes our choice appropriate. Finally, we are not accounting for glacial isostatic adjustment. found that the isostatic adjustment was not critical for glacial inception, though including the adjustment revealed a small positive feedback. Therefore, contrary to conventional wisdom, these studies suggest that isostatic adjustment might actually favor glacial inception over Baffin Island, and may even be critical to the process by reducing the strength of negative elevation–circulation feedbacks on ice growth (as in the “no H” sensitivity test). Our binning procedure and the resulting surface mass balance (SMB) forcing routine, while simple, are consistent with the way many simple ice sheet models are forced. More importantly, the results of this procedure, showing that ice elevation causes a negative warming feedback, should be robust regardless of how the ice elevation was calculated, as it is a result of the atmospheric model itself. We found that finer bin sizes did not significantly affect the expansion of ice. However, it would be good to test alternative SMB routines, such as using mean monthly air temperatures and precipitation from the previous WRF integrations, and downscale them to the surface elevation of ice grid points.

Though we find negative feedbacks on the growth of ice sheets due to clouds and circulation preventing inception in our model, high-resolution regional modeling still has the potential to make progress on the inception problem. We have captured the mass balance on critical mountainous regions in our Iteration 1 run, which GCMs cannot simulate accurately. When combined with ice flow, the Iteration 1 mass balance causes a significant portion of Baffin Island and the surrounding area to become ice covered. Though the equivalent sea level drop is only 6 m in 10 kyrs, the geological constraints on sea level drop from the Laurentide are not tight, and the calculated sea level drop may be consistent with more recent sea level estimates . However, this scenario neglects the ice–elevation feedback on the atmosphere that we found significant. Given the geological importance of Baffin Island on glacial inception, additional work is needed to understand how Baffin Island cooled against two warming influences: the regional anticyclone circulation and increasing insolation from 115 kya onward. The warming feedback due to local circulation changes that we find could be mitigated by changes in the pattern of large-scale atmospheric circulation, including stationary and planetary waves. Using a circumpolar regional model could potentially capture the mass balance of the nucleation points of both the Laurentide and Fennoscandian ice sheets at a sufficiently high resolution, allowing an investigation in the downstream interplay of these two ice sheets. Another interesting study could involve exploring the possible impact of a smaller Greenland ice sheet upon exiting the last interglacial. Going a step further to more non-local feedbacks (such as SST changes) influenced by changing orbital parameters, a next step would be simulating the last glacial inception with a GCM and using the results as boundary conditions in a regional model.